This is only affecting the nova-compute service running the libvirt driver. (We can ignore ironic computes)

Greenfield

Generate a random uuid and put into the compute uuid file that it is stored in the hypervisor and visible but read only in nova_compute container under /etc/nova. This will ensure that nova-compute will not generate a new uuid but will use the pre-generated one.

The upstream feature is available in Antelope so this part can be done right away.

Adoption

When adopting from 17.1 there will be a uuid file already generated by the running compute service (downstrem backport pending https://code.engineering.redhat.com/gerrit/q/topic:stable-compute-uuid ) . So during adoption the generated uuid file need to be preserved. Also to keep the greenfield and adopted deployments the same move the file from /var/lib/nova/compute_id to /etc/nova and make it read only for the nova_compute container.